A survey of view selection methods
Imene Mami, Zohra Bellahsene
To cite this version:
Imene Mami, Zohra Bellahsene. A survey of view selection methods. SIGMOD record, ACM,
2012, 41 (1), pp.20-29. .
HAL Id: lirmm-00720157
https://hal-lirmm.ccsd.cnrs.fr/lirmm-00720157
Submitted on 23 Jul 2012
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diffusion de documents
scientifiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
A Survey of View Selection Methods
Imene Mami and Zohra Bellahsene
Université Montpellier 2 - INRIA, LIRMM, Montpellier, France
{imen.mami,[email protected]}
ABSTRACT
Materialized view selection is a critical problem in many
applications such as query processing, data warehousing, distributed and semantic web databases, etc. We
refer to the problem of selecting an appropriate set of
materialized views as the view selection problem. Many
different view selection methods have been proposed in
the literature to address this issue. The present paper
provides a survey of view selection methods. It defines
a framework for highlighting the view selection problem
by identifying the main dimensions that are the basis in
the classification of view selection methods. Based on
this classification, this study reviews most of the view
selection methods by identifying respective potentials
and limits.
1.
INTRODUCTION
The view selection issue has been investigated in
several contexts: query optimization, warehouse design, data placement in a distributed setting, web
databases, etc. Many diverse solutions to the view
selection problem have been proposed and analyzed
through surveys [13, 21, 32]. The survey [21] concentrates on methods of finding a rewriting of a
query using a set of materialized views. The study
presented in [32] focuses on the state of the art in
materialization for web databases. A critical analysis of methodologies for selecting materialized views
in data warehousing is provided in [13]. However,
none of the above mentioned surveys provides a
classification of view selection approaches in order
to identify their advantages and disadvantages. Our
survey fills this gap.
This paper aims at studying the view selection in
relational databases and data warehouses as well as
in a distributed setting. First, we define a framework for highlighting the view selection problem.
Thus, we present a classification of view selection
methods based on the main view selection dimensions that we have identified. This study also reviews existing view selection methods by identifying
respective potentials and limits.
The rest of the paper is organized as follows: Section 2 gives some definitions. Section 3 identifies the
main view selection dimensions along which view selection methods can be classified. Section 4 presents
a critical survey of existing view selection methods.
Section 5 contains the conclusion and discusses open
issues.
2.
2.1
PROBLEM SPECIFICATION
Preliminaries
Here, we introduce the main notions used in this
paper and related to the view selection context.
View: A view is a derived relation, defined by a
query in terms of base relations and/or other views.
Materialized View: A view is said to be materialized if its query result is persistently stored
otherwise it is said to be virtual. We refer to a set
of selected views to materialize as a set of materialized views.
Workload: A workload or a query workload is
a given set of queries Q = {Q1 , Q2 , · · · , Qq }. Each
query Qi has an associated non-negative weight f Qi
which describes the query frequency. The set of materialized views is dependent on the query workload.
In a distributed scenario, the queries are executed
on different computer nodes. Each computer node
has an associated query workload.
View Selection: Given a database schema and
a query workload, the objective is to select an appropriate set of materialized views to improve query
performance. The process of selecting a set of materialized views is known as view selection.
View Maintenance: Whenever a base relation
is changed, the materialized views built on it have
to be updated in order to compute up-to-date query
results. The process of updating a materialized view
is known as view maintenance. Different maintenance policies (deferred or immediate) and maintenance strategies (incremental or rematerialization)
can be applied [17, 18, 38, 58].
of them.
2.2
2.3
Problem Definition
The use of materialized views is a common technique to reduce query response time [6]. Indeed,
materializing an appropriate set of views and answering queries using these views can significantly
speed up the query processing since the access to
materialized views can be much faster than recomputing the views. Therefore, materializing all the
input queries can achieve the lowest query processing cost but the highest view maintenance cost since
materialized views have to be maintained in order
to keep them consistent with the data at sources.
Besides, the query result can be too large to fit in
the available storage space. Hence, there is a need
for selecting a set of views to materialize by taking into account three important parameters: query
processing cost, view maintenance cost and storage space. The problem of choosing which views to
materialize which have a desirable balance among
the three costs is known as the view selection problem. This is one of the most challenging problems in
data warehousing [50] and it is known to be a NPcomplete problem [26]. In a distributed environment consisting of many heterogeneous nodes with
different resource constraints, the distributed view
selection problem is that which view has to be materialized at which node of the network. The view
selection problem in a distributed case is much more
difficult than the view selection problem in a central
case because of the immense challenges associated
to distributed settings [16] (i.e., data granularity,
degrees of replication, heterogeneity of information
sources, etc.).
Problem Formulation: The problem of view
selection can be formulated as follows. Given a
database schema R = {R1 , R2 , · · · , Rr }, a query
workload Q = {Q1 , Q2 , · · · , Qq } defined over R, the
problem is to select an appropriate set of materialized views M = {V1 , V2 , · · · , Vm } such that the
query workload is answered with the lowest cost
under a limited amount of resources, e.g., storage
space and/or view maintenance cost.
The view selection problem in a distributed context consisting of a set of nodes N = {N1 , N2 , · · · , Nn }
in which each node has an associated query workload, is to choose a set of views M = {V1 , V2 , · · · ,
Vm } and a set of nodes Nv ⊆ N at which M should
be materialized. The distributed view selection is
designed so that the full query workload is answered
with the lowest cost subject to resource constraints.
Resources may be storage space per node, view maintenance cost, communication costs or a combination
Cost Model
The cost model is an important issue for the view
selection process [9]. The main objective in view selection problem is the minimization of the weighted
query processing cost, defined by the formula:
X
QueryP rocessingCost =
fQi ∗ Qc(Qi , M )
Qi ∈Q
where fQi is the query frequency of the query Qi
and Qc(Qi , M ) is the processing cost corresponding to Qi given a set of materialized views M .
Because materialized views have to be kept up to
date, the view maintenance cost has to be considered. This cost is weighted by the update frequency
indicating the frequency of updating materialized
views. The view maintenance cost is computed as
follows:
X
V iewM aintenanceCost =
fu (Vi ) ∗ M c(Vi , M )
Vi ∈M
where fu (Vi ) is the update frequency of the view Vi
and M c(Vi , M ) is the maintenance cost of Vi given
a set of materialized views M .
The cost model is extended for distributed setting by taking into account the communication cost
which is the cost for transferring data from its origin
to the node that initiated the query. Given a query
Qi which is asked at a node Nj and denoting by Vk
a view used to answer Qi , the communication cost
is zero if Vk is materialized at Nj . Otherwise, let Nl
be the node containing Vk , then the communication
cost for transferring Vk from Nl to Nj is:
CommunicationCost(Vk ,Nl →Nj ) = CNj ,Nl ∗size(Vk )
where CNj ,Nl is the network transmission cost per
unit of data transferred between Nj and Nl and
size(Vk ) is the size of the view Vk .
2.4
Static View Selection vs. Dynamic View
Selection
A static view selection approach is based on a
given workload and chooses accordingly the set of
views to materialize. Whereas, in a dynamic view
selection approach, the view selection is applied as
a query arrives. Therefore, the workload is built
incrementally and changes over time. Because the
view selection has to be in synchronization with the
workload, any change to the workload should be reflected to the view selection as well. Indeed, in a
system of a dynamic nature [4, 5, 29], the set of
materialized views can be changed over time and replaced with more beneficial views in case of changing the query workload. In order to reduce view
maintenance cost and storage space requirements,
[57] aims at materializing the most frequently accessed tuples of the view rather than materializing all tuples of the view. The set of materialized
tuples can be changed dynamically as the queries
change, either manually or automatically by an internal cache manager using a feedback loop. However, the task of monitoring constantly the query
pattern and periodically recalibrating the materialized views is rather complicated and time consuming especially in large data warehouse where many
users with different profiles submit their queries.
A dynamic view selection is often referred to as
view caching. With caching, the cache is initially
empty and data are inserted or deleted from the
cache during the query processing. Materialization
could be performed even if no queries have been
processed and materialized views have to be updated in response of changes on the base relations.
A detailed comparison of these two techniques is
given in [27]. Traditional caching approaches aim
at caching the results of queries, in other words to
cache views. Another alternative is to cache only a
part of a view. Indeed, a chunk based scheme has
been introduced in [12] for fine granularity caching.
Chunk based caching allows caching of only few,
frequently used tuples of views. To facilitate the
computation of chunks required by a query but not
found in the cache, a new organization for base relations has been proposed which they called a chunked file. Caching has been adopted in data warehousing [43], distributed databases [28] and peer to
peer systems [25]. Dynamic view indexing has also
been considered in [44]. In this paper, we focus
only on static view selection methods because most
of existing view selection approaches are of static
nature.
3.
VIEW SELECTION DIMENSIONS
In order to identify the advantages and disadvantages of view selection methods, we propose two
main dimensions along with they can be classified:
(i) Frameworks; and (ii) Resource Constraints.
3.1
Frameworks
Generally, approaches to the view selection problem consist of two main steps. The first step identifies the candidate views which are promising for
materialization. Techniques based on multiquery
DAG, syntactical analysis of the workload or query
rewriting have been used to obtain the candidate
Figure 1: The AND-OR view graph of the
two queries Q1 and Q2 .
views. Based on the set of candidate views, the second step selects the set of views to materialize under
the resource constraints and by applying heuristic
algorithms.
3.1.1
Multiquery DAG
Most of the proposed view selection methods operate on query execution plans. The plans can be
derived from multiple query optimization techniques
or by merging multiple query plans. The main interest of such techniques relies in detecting common sub-expressions between the different queries
of workload and capturing the dependencies among
them. This feature can be exploited for sharing updates and storage space. The dependence relation
on queries (or views) has been represented by using
a directed acyclic graph also called a DAG. However, these methods require optimizer calls which
can be expensive in complex scenarios.
The most commonly used DAGs in literature are:
AND/OR View Graph: The union of all possible execution plans of each query forms an ANDOR view graph [40]. The AND-OR view graph described by Roy [42] is a Directed Acyclic Graph
(DAG) composed of two types of nodes: Operation
nodes and Equivalence nodes. Each operation node
represents an algebraic expression (Select-ProjectJoin) with possible aggregate function. An equivalence node represents a set of logical expressions
that are equivalent (i.e., that yield the same result).
The operation nodes have only equivalence nodes as
children and equivalence nodes have only operation
nodes as children. The root nodes are the query results and the leaf nodes represent the base relations.
A sample AND-OR view graph is shown in figure
1. Circles represent operation nodes (Op-Nodes)
and boxes represent equivalence nodes (Eq-Nodes).
For example, in figure 1, view V1 corresponding to
a single query Q1 , can be computed from V6 and
V3 or R1 and V4 . If there is only one way to answer or update a given query, the graph becomes
an AND view graph. In the data cube which is a
specific model of a data warehouse, the AND-OR
view graph is an OR view graph, as for each view
there are zero or more ways to construct it from
other views, but each way involves only one other
view [19]. In other words, an OR view graph is an
AND-OR view graph in which any node is an equivalence node that can be computed from any one of
its children.
Multi-View Processing Plan (MVPP): The
MVPP defined by Yang et al [52] is a directed acyclic
graph in which the root nodes are the queries, the
leaf nodes are the base relations and all other intermediate nodes are selection, projection, join or
aggregation views that contribute to the construction of a given query. The MVPP is obtained after
merging into a single plan either individual optimal query plans (similar to the AND view graph)
or all possible plans for each query (similar to the
AND-OR view graph). The difference between the
MVPP representation and the AND-OR view graph
or the AND view graph representation is that all intermediate nodes in the MVPP represent operation
nodes. A sample MVPP is shown in figure 2.
Data Cube Lattice: Harinarayan and al [22]
propose modeling data in multiple dimensions. It
is built from the queries involved in the data warehouse application, e.g., OLAP-style queries. The
Data Cube Lattice is a DAG whose nodes represent
queries (or views) which are characterized by the
attributes of the Group by clause. The edges denote the derivability relation between views. That
is, if there is a path from view Vi to a view Vj (see
figure 3), then grouping attributes on Vj can be calculated from grouping attributes on Vi . The node
labeled none corresponds to an empty set of groupby attributes (tuples are not grouped). The benefit
of this representation is that a query can be used
to answer or update another query. An extension
of the data cube lattice in order to adapt it to a
distributed case was proposed in [3, 53]. Indeed,
the cube has been modified by adding edges that
mark the derivation relationship between views on
different computer nodes.
Figure 2: The MVPP of the two queries Q1
and Q2 .
Figure 3: Sample Data Cube Lattice for eight
views.
3.1.2
Query Rewriting
Query rewriting based approaches not only compute the set of materialized views but also find a
complete rewriting of the queries over it. Here, the
input to the view selection is not a multiquery DAG
but the query definitions. The view selection problem is modeled as a state search problem using a
set of transformation rules. These rules detect and
exploit common subexpressions between the queries
of the workload and guarantee that all the queries
can be answered using exclusively the materialized
views. Query rewriting based approaches not only
compute the set of materialized views but also find
a complete rewriting of the queries over it. Nevertheless, the completeness of the transformation
rules may make the complexity of state space search
strategies exponential.
3.1.3
Syntactical Analysis of the Workload
Some view selection methods are based on syntactical analysis of the workload to identify candidate
views. These approaches analyze the workload and
pick a subset of relations from which to materialize
one or more views, if only if has the potential to
reduce the cost of the workload significantly. How-
ever, the search space for computing the optimal set
of views to be materialized may be very large.
3.2
Resource Constraints
Resource constraints considered during the view
selection can be taken into account when classifying view selection methods. There are three main
models presented in literature: unbounded, space
constrained and maintenance cost constrained.
3.2.1
Unbounded
In the unbounded setting, there is no limit on
available resources (storage, computation etc.). Thus,
the view selection problem consists in choosing a
set of views to materialize that minimizes the query
processing cost and the view maintenance cost. Formally thus, the problem is:
P
minimize
Qi ∈Q fQi ∗ Qc(Qi , M )
P
+ Vi ∈M fu (Vi ) ∗ M c(Vi , M )
However, this approach may lead to two kinds
of problems. First, sometimes the selected views
may be too large to fit in the available space. Second, the cost of the view maintenance may offset
the performance advantages provided by the view
materialization.
3.2.2
Space Constrained
Due to the storage space limitation, materializing
all views is not always possible. In this setting, a
useful notion is that of a view benefit (or query benefit). This is defined as the reduction in the workload evaluation cost, that can be achieved by materializing this view. Also relevant in this context is
the per-unit benefit, obtained by dividing the view
benefit by its space occupancy. It has been shown
[19] that the per-space unit benefit of a view can
only decrease as more views are selected (monotonic
property). The space constrained model minimizes
the query processing cost plus the view maintenance
cost under a space constraint.
P
minimize
Qi ∈Q fQi ∗ Qc(Qi , M )
P
+ Vi ∈M fu (Vi ) ∗ M c(Vi , M )
under
P
Vi ∈M
size(Vi ) ≤ S
where S is the storage space capacity.
3.2.3
Maintenance Cost Constrained
This model constrains the time that can be allotted to keep up to date the materialized views in
response to updates on base relations. In the maintenance cost constrained model, the maintenance
cost of a view may decrease with selection of other
views for materialization. Therefore, the query benefit per unit of maintenance cost of a view can increase [20]. This non monotonic nature of maintenance cost makes the view selection problem more
difficult. The maintenance cost constrained model
minimizes the query processing cost under a maintenance cost constraint.
P
minimize
Qi ∈Q fQi ∗ Qc(Qi , M )
under
P
Vi ∈M
fu (Vi ) ∗ M c(Vi , M ) ≤ U
where U is the view maintenance cost limit.
The models that we have presented in section
3.2 can be extended to the distributed setting by
taking into account the distributed specific features
(i.e., the communication cost between the computer
nodes).
4.
REVIEW OF VIEW SELECTION METHODS
In this section, we classify the view selection methods according to several dimensions characterizing
their algorithms (i) resource constraints they consider during the view selection process and (ii) frameworks they use to obtain the candidate views (see
figure 4). Based on this classification, we review
most of the view selection methods. The best-known
heuristic algorithms proposed in litterature to solve
the view selection problem, namely: deterministic algorithms, randomized algorithms, hybrid algorithms or constraint programming.
4.1
Deterministic Algorithms Based Methods
Much research work on view selection uses deterministic strategies to address the view selection
problem. [41] is the first paper that provides a solution for materializing view indexes which can be
seen as a special case of the materialized views. The
solution is based on A* algorithm [37]. An exhaustive approach is also presented in [31, 39] for finding
the best set of views to materialize. Nevertheless,
an exhaustive search cannot compute the optimal
solution in a reasonable time.
The authors in [22] present and analyze algorithms for view selection in case of OLAP-style queries.
They provide a polynomial-time greedy algorithm
to select a set of views to materialize that minimizes the query processing cost subject to a space
constraint. However, this approach does not con-
Figure 4: A Classification of view selection methods.
sider the view maintenance cost. The work in [51]
is dealing with more general SQL queries which include select, project, join, and aggregation operations. A greedy algorithm has been designed to
select a set of materialized views so that the combined query processing and view maintenance cost
is minimized. However, the view maintenance cost
has been overrated since the maintenance cost for
a materialized view is the cost used for constructing this view. Besides, the view selection is done
without any resource constraint.
A theoretical framework for the view selection
problem in data warehousing setting has been developed in [19]. Their work provides a near-optimal
exponential time greedy algorithm for the case of
AND-OR view graph and near-optimal polynomial
time greedy algorithm for the cases of AND view
graph and OR view graph. This approach was extended in [20] to study the view selection under a
maintenance cost constraint.
The authors in [42] demonstrate that using multiquery optimization techniques in conjunction with a
greedy heuristic is practical and provides significant
benefit. The greedy heuristic is used to iteratively
pick from the AND-OR view graph the set of views
to materialize that minimizes the query processing
cost. This study was extended in [36] to consider
how to optimize view maintenance cost. In addition to speed up the query workload by selecting
materialized views, algorithms exploit common subexpressions between view maintenance expressions
to compute an efficient plan to the maintenance of
the materialized views. However, the view selection
has been studied without any resource constraint.
The view selection algorithm proposed in [2] is
based on the notion of level in the query tree (each
view of the query tree is associated to a level). In
this approach, the view selection problem is studied
under a space constraint and solved in two phases.
The first phase depends on local optimization by
taking each query and pre-selecting a set of views
which reduce the query processing cost without increasing significantly the view maintenance cost.
The second phase computes the cost for each level
of the query graph and selects the one which has the
minimal sum of query processing and view maintenance cost.
The view selection has been studied in [34, 45, 46,
47, 48] under the condition that the input queries
can be answered using exclusively the materialized
views. An exhaustive algorithm has been designed
in [47] to select a set of materialized views while
minimizing the combination of the query processing and view maintenance cost. This work was extended in [34] by developing greedy algorithms that
expand only a small fraction of the states produced
by the exhaustive algorithm. The view selection
problem in [45, 46, 48] is addressed under a space
constraint. However, their view selection algorithm
is still in exponential time. A survey of work on
answering queries using views can be found in [21].
The study in [1] is based on a syntactical analysis
of the workload to address the problem of selecting
both views and indexes to be materialized. This
approach proceeds in three main steps. The first
step analyses the workload and chooses subsets of
base relations with a high impact on the query processing cost. Based on the base relations subsets,
the second step identify syntactically relevant views
and indexes that can potentially be materialized. In
the third step, the system runs a greedy enumeration algorithm to pick a set of views and indexes to
materialize based on the result of the second step
by taking into account the space constraint. Nevertheless, this approach does not take into account
the view maintenance cost.
The works published in [3, 53] address the view
selection problem in a distributed data warehouse
environment. An extension of the concept of a data
cube lattice to capture the distributed semantics
has been proposed. Moreover, they extend a greedy
based selection algorithm for the distributed case.
However, the cost model that they have used does
not include the view maintenance cost. Furthermore, the network transmission costs are not considered which is very important in a distributed context. Indeed, the communication cost is computed
only from the size of the query result.
The above methods take a deterministic approach
either by exhaustive search or by some heuristics
such as greedy. However, greedy search is subjected to the known caveats, i.e., sub-optimal solutions may be retained instead of the globally optimal one since initial solutions influence the solu-
tion greatly. As a result, many paradigms and programming techniques have been developed to improve the solutions of the view selection problem:
randomized algorithms, hybrid algorithms and constraint programming which we describe in next subsection.
4.2
Randomized Algorithms Based Methods
Typical randomized algorithms are genetic [14]
or use simulated annealing [30]. Genetic algorithms
generate solutions using techniques inspired by the
natural evolution process such as selection, mutation, and crossover. The search strategy for these
algorithms is very similar to biological evolution.
Genetic algorithms start with a random initial population and generate new populations by random
crossover and mutation. The fittest individual found
is the solution. The algorithms terminate as soon
as there is no further improvement over a period.
A genetic algorithm has been used in [23, 55]
in conjunction with the MVPP framework to solve
the view selection problem. The materialized views
have been selected according to their reduction in
the combined query processing and view maintenance cost. However, because of the random characteristic of the genetic algorithm, some solutions
can be infeasible. For example, in the maintenance
cost constrained model, when a view is selected, the
benefit will not only depend on the view itself but
also on other views that are selected. One solution
to this problem is to add a penalty value as part
of the fitness function to ensure that infeasible solutions will be discarded. For instance, a penalty
function has been applied in [33] which reduces the
fitness each time the maintenance cost constraint is
not satisfied. This approach minimizes the query
processing cost given varying upper bounds on the
view maintenance cost, assuming unlimited amount
of storage space. In order to let the genetic algorithm converge faster, they represent the initial
population as a favorable configuration based on
external knowledge about the problem and its solution rather than a random sampling, i.e., the views
with a high query frequency are most likely selected
for materialization. However, the genetic algorithm
may tend to get stuck at a poor local optimum fairly
early. A solution was provided in [54] to avoid premature convergence and keep improving the solution by incorporating constraints into the algorithm
through a stochastic ranking procedure where no
penalty functions are used.
The study presented in [8] which is based on a
syntactical analysis of the workload deals with the
distributed view selection problem. This approach
consists of three main steps. The first one extends
the base relations selection algorithm described in
[1] for the distributed scenario. Based on the result
of the first step and the similarity between queries,
the second step generates the candidate views which
are promising for materialization. In the third step
a genetic algorithm is applied to select a set of materialized views and the nodes of the network on
which they will be materialized that minimize the
query processing and view maintenance cost. However, this approach does not take into account either the space constraint or the maintenance cost
constraint.
The approaches proposed in [10, 11, 24] use simulated annealing algorithms to address the view selection problem. These algorithms are motivated
by an analogy to annealing in solids. Simulated
Annealing algorithms start with an initial configuration, generate new configurations by random walk
along the different solutions of the solution space according to a cooling schedule and terminate as soon
as no applicable ones exist or lose all the energy in
the system.
Materialized views have been selected in [10] so
that the combined query processing and view maintenance cost is minimized. The view selection problem is solved in [24] under the case where either the
space constraint or the maintenance cost constraint
is considered. Further, randomized search has been
applied to solve two more issues. First, they considered the case where both space and maintenance
constraints exist. Next they applied randomized
search in the context of dynamic view selection.
In order to support the scalability when the number of views and queries become large, a new approach has been introduced in [11] using Parallel
Simulated Annealing (PSA) for materialized view
selection. By performing simulated annealing with
multiple inputs over multiple compute nodes concurrently, PSA is able to improve the quality of obtained sets of materialized views. Moreover, PSA
is able to perform view selection on MVPP having
a much larger number of views, which reflects the
real data warehousing environment. However, the
view selection problem is solved without any bound
neither on the storage space nor on the view maintenance cost.
Randomized algorithms can be applied to complex problems dealing with large or even unlimited
search spaces. Thus, the use of randomized algorithms can be considered in solving large combinatorial problems such as the view selection problem.
However, the quality of the solution 1 depends on
the set-up of the algorithm as well as the extremely
difficult fine-tuning of algorithm that must be performed during many test runs.
4.3
Hybrid Algorithms Based Methods
Hybrid algorithms combine the strategies of deterministic and randomized algorithms in their search
in order to provide better performance in terms of
solution quality. Solutions obtained by deterministic algorithms are used as initial configuration for
simulated annealing algorithms or as initial population for genetic algorithms.
A hybrid approach has been applied in [56] which
combines heuristic algorithms i.e., greedy algorithms
and genetic algorithms to solve three related problems. The first one is to optimize queries. The second one is to choose the best global processing plan
from multiple processing plans for each query. The
third problem is to select materialized views from
a given global processing plan. Their experimental results confirmed that hybrid algorithms provide
better performance than either genetic algorithms
or heuristic algorithms i.e., greedy algorithms used
alone in terms of solution quality. However, their algorithms are more time consuming and may be impractical due to their excessive computation time.
4.4
Constraint Programming Based Methods
Constraint programming is a descendant of declarative programming. This programming technique
has been exploited in many applications for solving combinatorial problems [49]. The success of using constraint programming for combinatorial optimization is due to its combination of high level
modeling, constraint propagation and facilities to
control the search behavior.
A constraint programming based approach has
been presented in [35] to address the view selection
problem. More specifically, the view selection problem has been modeled as a constraint satisfaction
problem. Its resolution has been supported automatically by constraint solver embedded in the constraint programming language. The authors proved
experimentally that a constraint programming based
approach provides better performances compared
with a randomized method i.e., genetic algorithm
in term of cost savings. The view selection has been
studied under the case where (i) only the maintenance cost constraint is considered and (ii) both
1
The solution quality represents the quality of the set
of materialized views found by the algorithm. For example, the solution quality may be measured in term of
cost savings.
maintenance cost and space constraints exist. They
have also shown that their approach is scalable.
5.
CONCLUSION
This study provides a critical survey of different
approaches in which the view selection has been
studied in relational databases and data warehouses
as well as in a distributed setting. We have defined
formally the view selection problem and identified
the main view selection dimensions along with view
selection methods have been classified. Based on
the classification, we have discussed most of view
selection methods.
Analysis of state of the art of view selection has
shown that there is very few work on view selection
in distributed databases and data warehouses [3,
8, 53] and no effective solution for peer to peer systems. Indeed, [16] seems to be the only paper which
deals with the view selection problem in peer to peer
environment. In fact, it is provided a full definition
of the problem but without providing any algorithm
or detail on how to select an effective set of views
to materialize and place them at appropriate peers.
Thus, one of challenging directions of future work
aims at addressing the view selection problem in
a distributed setting. More recently, materialized
view selection has been explored in semantic web
databases [7, 15] in order to facilitate efficient processing of RDF queries and updates. However, they
consider a static workload which contradicts the dynamic nature of the web. Indeed, any change to
the workload should be reflected to the view selection as well. This issue will be the future aspect
while studying the view selection in semantic web
databases.
6.
ACKNOWLEDGEMENTS
We would like to thank the reviewers for their
valuable comments to improve this paper.
7.
REFERENCES
[1] S. Agrawal, S. Chaudhuri, and V.R. Narasayya.
Automated selection of materialized views and indexes
in SQL databases. In VLDB, pages 496–505, 2000.
[2] X. Baril and Z. Bellahsene. Selection of materialized
views: A cost-based approach. In CAiSE, pages
665–680, 2003.
[3] A. Bauer and W. Lehner. On solving the view selection
problem in distributed data warehouse architectures.
In SSDBM, pages 43–, 2003.
[4] Z. Bellahsene, R. Coenmans, and J. Tranier.
Matérialisation de vues dans les entrepôts de données.
une approche dynamique. Ingénierie des Systèmes
d’Information, 11(6):33–53, 2006.
[5] Z. Bellahsene, M. Cart, and N. Kadi. A cooperative
approach to view selection and placement in P2P
systems. In OTM, pages 515–522, 2010.
[6] R.G. Bello, K. Dias, A. Downing, J. Feenan, J.L.
Finnerty, W.D. Norcott, H. Sun, A. Witkowski, and
M. Ziauddin. Materialized views in ORACLE. In
VLDB, pages 659–664, 1998.
[7] R. Castillo, and U. Leser. Selecting materialized views
for RDF data. In ICWE Workshops, 2010.
[8] L.W.F. Chaves, E. Buchmann, F. Hueske, and K.
Böhm. Towards materialized view selection for
distributed databases. In EDBT, pages 1088–1099,
New York, NY, USA, 2009. ACM.
[9] R. Chirkova, A.Y. Halevy, and D. Suciu. A formal
perspective on the view selection problem. VLDB J.,
11(3):216–237, 2002.
[10] R. Derakhshan, F.K.H.A. Dehne, O. Korn, and B.
Stantic. Simulated annealing for materialized view
selection in data warehousing environment. In
Databases and Applications, pages 89–94, 2006.
[11] R. Derakhshan, B. Stantic, O. Korn, and F.K.H.A
.Dehne. Parallel simulated annealing for materialized
view selection in data warehousing environments. In
ICA3PP, pages 121–132, 2008.
[12] P. Deshpande, K. Ramasamy, A. Shukla, and J.F.
Naughton. Caching multidimensional queries using
chunks. In SIGMOD Conference, pages 259–270, 1998.
[13] C.A. Dhote, and M.S. Ali. Materialized View Selection
in Data Warehousing: A Survey. In Journal of Applied
Sciences, pages 401–414, 2009.
[14] D.E. Goldberg. Genetic Algorithms in Search
Optimization and Machine Learning. Addison-Wesley,
1989.
[15] F. Goasdoue, K. Karanasos, J. Leblay, and I.
Manolescu. View Selection in Semantic Web
Databases. In PVLDB, pages 97–108, 2011.
[16] S.D. Gribble, A.Y. Halevy, Z.G. Ives, M. Rodrig, and
D. Suciu. What can database do for peer-to-peer? In
WebDB, pages 31–36, 2001.
[17] A. Gupta and I.S. Mumick. Maintenance of
materialized views: Problems, techniques, and
applications. IEEE Data Eng. Bull., 18(2):3–18, 1995.
[18] A. Gupta, I.S. Mumick, and V.S. Subrahmanian.
Maintaining views incrementally. In SIGMOD
Conference, pages 157–166, 1993.
[19] H. Gupta. Selection of views to materialize in a data
warehouse. In ICDT, pages 98–112, 1997.
[20] H. Gupta and I.S. Mumick. Selection of views to
materialize under a maintenance cost constraint. In
ICDT, pages 453–470, 1999.
[21] A.Y. Halevy. Answering queries using views: A survey.
VLDB J., 10(4):270–294, 2001.
[22] V. Harinarayan, A. Rajaraman, and J.D. Ullman.
Implementing data cubes efficiently. In SIGMOD
Conference, pages 205–216, 1996.
[23] J.T. Horng, Y.J. Chang, and B.J. Liu. Applying
evolutionary algorithms to materialized view selection
in a data warehouse. Soft Comput., 7(8):574–581, 2003.
[24] P. Kalnis, N. Mamoulis, and D. Papadias. View
selection using randomized search. Data Knowl. Eng.,
42(1):89–111, 2002.
[25] P. Kalnis, W.S. Ng, B.C. Ooi, D. Papadias, and K.L.
Tan. An adaptive peer-to-peer network for distributed
caching of OLAP results. In SIGMOD Conference,
pages 25–36, 2002.
[26] H.J. Karloff and M. Mihail. On the complexity of the
view-selection problem. In PODS, pages 167–173, 1999.
[27] D. Kossmann. The state of the art in distributed query
processing. ACM Comput. Surv., 32(4):422–469, 2000.
[28] D. Kossmann, M.J. Franklin, and G. Drasch. Cache
investment: integrating query optimization and
distributed data placement. ACM Trans. Database
Syst., 517–558, 2000.
[29] Y. Kotidis and N. Roussopoulos. Dynamat: A dynamic
view management system for data warehouses. In
SIGMOD Conference, pages 371–382, 1999.
[30] P.J.M. Laarhoven and E.H.L. Aarts, editors. Simulated
annealing: theory and applications. Kluwer Academic
Publishers, Norwell, MA, USA, 1987.
[31] W. Labio, D. Quass, and B. Adelberg. Physical
database design for data warehouses. In ICDE, pages
277–288, Washington, DC, USA, 1997. IEEE
Computer Society.
[32] A. Labrinidis , Q. Luo , J. Xu, and W. Xue. Caching
and Materialization for Web Databases. In
Foundations and Trends in Databases, pages 169–266,
2009.
[33] M. Lee and J. Hammer. Speeding up materialized view
selection in data warehouses using a randomized
algorithm. Int. J. Cooperative Inf. Syst.,
10(3):327–353, 2001.
[34] S. Ligoudistianos, D. Theodoratos, and T.K. Sellis.
Experimental evaluation of data warehouse
configuration algorithms. In DEXA Workshop, pages
218–223, 1998.
[35] I. Mami, R. Coletta, and Z. Bellahsene. Modeling view
selection as a constraint satisfaction problem. In
DEXA, pages 396–410, 2011.
[36] H. Mistry, P. Roy, S. Sudarshan, and K.
Ramamritham. Materialized view selection and
maintenance using multi-query optimization. In
SIGMOD Conference, pages 307–318, 2001.
[37] N.J. Nilsson. Problem-Solving Methods in Artificial
Intelligence. McGraw-Hill Pub. Co., 1971.
[38] D. Quass, A. Gupta, I.S. Mumick, and J. Widom.
Making views self-maintainable for data warehousing.
In PDIS, pages 158–169, 1996.
[39] K.A. Ross, D. Srivastava, and S. Sudarshan.
Materialized view maintenance and integrity constraint
checking: Trading space for time. In SIGMOD
Conference, pages 447–458, 1996.
[40] N. Roussopoulos. The logical access path schema of a
database. IEEE Trans. Software Eng., 8(6):563–573,
1982.
[41] N. Roussopoulos. View indexing in relational
databases. ACM Trans. Database Syst., 7(2):258–290,
1982.
[42] P. Roy, S. Seshadri, S. Sudarshan, and S. Bhobe.
Efficient and extensible algorithms for multi query
optimization. In SIGMOD Conference, pages 249–260,
2000.
[43] P. Scheuermann, J. Shim, and R. Vingralek.
Watchman : A data warehouse intelligent cache
manager. In VLDB, pages 51–62, 1996.
[44] K. Schnaitter, S. Abiteboul, T. Milo, and N. Polyzotis.
Colt: continuous on-line tuning. In SIGMOD
Conference, pages 793–795, 2006.
[45] D. Theodoratos, S. Ligoudistianos, and T.K. Sellis.
Designing the global data warehouse with SPJ views.
In CAiSE, pages 180–194, 1999.
[46] D. Theodoratos, S. Ligoudistianos, and T.K. Sellis.
View selection for designing the global data warehouse.
Data Knowl. Eng., 39(3):219–240, 2001.
[47] D. Theodoratos and T.K. Sellis. Data warehouse
configuration. In VLDB, pages 126–135, 1997.
[48] D. Theodoratos and T.K. Sellis. Data warehouse
schema and instance design. In ER, pages 363–376,
1998.
[49] M. Wallace. Practical applications of constraint
programming. Constraints, 1:139–168, 1996.
10.1007/BF00143881.
[50] J. Widom. Research problems in data warehousing. In
CIKM, pages 25–30, 1995.
[51] J. Yang, K. Karlapalem, and Q. Li. Algorithms for
materialized view design in data warehousing
environment. In VLDB, pages 136–145, 1997.
[52] J. Yang, K. Karlapalem, and Q. Li. A framework for
designing materialized views in data warehousing
environment. In ICDCS, 1997.
[53] W. Ye, N. Gu, G. Yang, and Z. Liu. Extended
derivation cube based view materialization selection in
distributed data warehouse. In WAIM, pages 245–256,
2005.
[54] J.X. Yu, X. Yao, C.H. Choi, and G. Gou. Materialized
view selection as constrained evolutionary
optimization. IEEE Transactions on Systems, Man,
and Cybernetics, Part C, 33(4):458–467, 2003.
[55] C. Zhang and J. Yang. Genetic algorithm for
materialized view selection in data warehouse
environments. In DaWaK, pages 116–125, 1999.
[56] C. Zhang, X. Yao, and J. Yang. An evolutionary
approach to materialized views selection in a data
warehouse environment. IEEE Transactions on
Systems, Man, and Cybernetics, Part C,
31(3):282–294, 2001.
[57] J. Zhou, P. Larson, J. Goldstein, and L. Ding. Dynamic
materialized views. In ICDE, pages 526–535, 2007.
[58] Y. Zhuge, H.G. Molina, J. Hammer, and J. Widom.
View maintenance in a warehousing environment. In
SIGMOD Conference, pages 316–327, 1995.